The EU-funded STAR-project provided an opportunity to analyse 1418 macroinvertebrate samples from 310 sampling sites throughout Europe. At most of the sites, samples were taken in two seasons using both national protocols and the project's STAR-AQEM protocol. At a subset of sites (86), two replicate samples were taken by each method in each of the two seasons. The resulting taxalists were analysed in terms of community similarity using the Bray-Curtis Index, Jaccard, and Renkonen Indices. A new concept of sample ‘coherence' is used to measure the relative strength of within-site, within-season and within-method similarity and to determine their importance on variability in community composition. Site-coherence (i.e., highest similarity to another sample from the same site) was much higher where replicate samples were available. Season-coherence of samples was nearly 100% even if different methods were compared. Season appeared to be one of the major determinants of in-stream fauna. The STAR-AQEM method is most comparable to the Nordic, Portuguese and Czech (PERLA) national methods and less comparable to the Italian (IBE) and Latvian methods. Samples collected by these latter methods had higher similarities to other sites sampled with the same methods than to samples of the same site using the STAR-AQEM method, thus there was low site-coherence. In three stream types from Italy, Latvia and Greece 28-38% of the samples were most similar to a sample from a different site than to a replicate sample from the same site. This fact could have serious consequences for follow up bioassessments or impact assessments by cluster analysis based on similarity measures. Replicate samples are less coherent within site, season or method if the taxonomic resolution is family rather than species.